Signal Detection Based On Neural Network

Signal detection is an important part of the radar system and the Constant False Alarm Rate(CFAR)detector occupies an indispensable position in the signal detection.Although the CFAR detection theory has made great progress,the existing classical CFAR detectors can only achieve better detection performance in specific clutter environments.For example,the Cell Averaging CFAR(CA-CFAR)detector has good detection performance in homogeneous clutter background,but the detection performance in multi-target and clutter edge is decreasing rapidly and so on.In this paper,based on the analysis of the traditional CFAR detector,two kinds of signal detectors based on neural network are proposed.The main work is as follows:(1)The basic principle of the traditional CFAR detector is discussed,and based on it,the actual clutter environment is built by simulation.The performance of several traditional CFAR detectors on homogeneous clutter,multi-target and clutter edges is discussed and compared.(2)A signal detection method called Neural Network and Constant False Alarm(NN-CFAR)is proposed by the combination of neural network and traditional constant false alarm method.The main idea of the detection method is to identify the clutter background(including homogeneous clutter,multi target and clutter edge environment)by using a multi-layer perceptron.Then select the appropriate detection methods(including single layer neural network,OS-CFAR and GO-CFAR detector)to realize the target recognition and detection.The optimal structure parameters and the training sample size of the neural network are determined by multiple training.Then the performance of the NN-CFAR signal detector is verified in different environment.Compared with the traditional CFAR detector,the NNCFAR signal detector has a strong ability to adapt to the environment,and the performance is relatively improved.(3)Based on the NN-CFAR signal detector,a signal detector called BMLP-TP signal detector is proposed which uses Multi-Layered Perception to achieve optimal threshold prediction in different clutter environment.The detector mainly uses the nonlinear characteristics of multilayer perceptron to realize the prediction of the traditional CFAR optimal threshold(including CA-CFAR,OS-CFAR and GO-CFAR threshold)for different clutter background(including homogeneous clutter,multi-target and clutter edge environment).After training for many times,the optimal structure parameters and the quantity of training samples of multilayer perceptron are determined,and then the performance of BMLP-TP signal detector in homogeneous clutter,multi-target and clutter edge is verified.The results show that BMLP-TP signal detector can be applied to a variety of clutter background and the performance is between the traditional CFAR and NN-CFAR signal detectors.(4)Based on the above two kinds of neural network signal detectors and a lot of data collected from real homogeneous clutter environment,the effectiveness of the proposed algorithm is verified.